Predictions represent the likelihood of something occurring. Unfortunately you cannot see the future, but based on the results of the predictive models you can foresee what will most likely happen.
When you evaluate your model you are in fact confirming whether the results you get are good enough so that you can trust what your model indicates. This prevents what is referred to as overfitting in Machine Learning. BigML allows you to evaluate all your training models with a simple three-click process so you can check if your model is able to make accurate predictions on data it has never seen before.